export.c 6.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255
  1. #include <linux/ceph/ceph_debug.h>
  2. #include <linux/exportfs.h>
  3. #include <linux/slab.h>
  4. #include <asm/unaligned.h>
  5. #include "super.h"
  6. #include "mds_client.h"
  7. /*
  8. * NFS export support
  9. *
  10. * NFS re-export of a ceph mount is, at present, only semireliable.
  11. * The basic issue is that the Ceph architectures doesn't lend itself
  12. * well to generating filehandles that will remain valid forever.
  13. *
  14. * So, we do our best. If you're lucky, your inode will be in the
  15. * client's cache. If it's not, and you have a connectable fh, then
  16. * the MDS server may be able to find it for you. Otherwise, you get
  17. * ESTALE.
  18. *
  19. * There are ways to this more reliable, but in the non-connectable fh
  20. * case, we won't every work perfectly, and in the connectable case,
  21. * some changes are needed on the MDS side to work better.
  22. */
  23. /*
  24. * Basic fh
  25. */
  26. struct ceph_nfs_fh {
  27. u64 ino;
  28. } __attribute__ ((packed));
  29. /*
  30. * Larger 'connectable' fh that includes parent ino and name hash.
  31. * Use this whenever possible, as it works more reliably.
  32. */
  33. struct ceph_nfs_confh {
  34. u64 ino, parent_ino;
  35. u32 parent_name_hash;
  36. } __attribute__ ((packed));
  37. static int ceph_encode_fh(struct dentry *dentry, u32 *rawfh, int *max_len,
  38. int connectable)
  39. {
  40. int type;
  41. struct ceph_nfs_fh *fh = (void *)rawfh;
  42. struct ceph_nfs_confh *cfh = (void *)rawfh;
  43. struct dentry *parent;
  44. struct inode *inode = dentry->d_inode;
  45. int connected_handle_length = sizeof(*cfh)/4;
  46. int handle_length = sizeof(*fh)/4;
  47. /* don't re-export snaps */
  48. if (ceph_snap(inode) != CEPH_NOSNAP)
  49. return -EINVAL;
  50. spin_lock(&dentry->d_lock);
  51. parent = dget(dentry->d_parent);
  52. spin_unlock(&dentry->d_lock);
  53. if (*max_len >= connected_handle_length) {
  54. dout("encode_fh %p connectable\n", dentry);
  55. cfh->ino = ceph_ino(dentry->d_inode);
  56. cfh->parent_ino = ceph_ino(parent->d_inode);
  57. cfh->parent_name_hash = ceph_dentry_hash(parent->d_inode,
  58. dentry);
  59. *max_len = connected_handle_length;
  60. type = 2;
  61. } else if (*max_len >= handle_length) {
  62. if (connectable) {
  63. *max_len = connected_handle_length;
  64. type = 255;
  65. } else {
  66. dout("encode_fh %p\n", dentry);
  67. fh->ino = ceph_ino(dentry->d_inode);
  68. *max_len = handle_length;
  69. type = 1;
  70. }
  71. } else {
  72. *max_len = handle_length;
  73. type = 255;
  74. }
  75. dput(parent);
  76. return type;
  77. }
  78. /*
  79. * convert regular fh to dentry
  80. *
  81. * FIXME: we should try harder by querying the mds for the ino.
  82. */
  83. static struct dentry *__fh_to_dentry(struct super_block *sb,
  84. struct ceph_nfs_fh *fh)
  85. {
  86. struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
  87. struct inode *inode;
  88. struct dentry *dentry;
  89. struct ceph_vino vino;
  90. int err;
  91. dout("__fh_to_dentry %llx\n", fh->ino);
  92. vino.ino = fh->ino;
  93. vino.snap = CEPH_NOSNAP;
  94. inode = ceph_find_inode(sb, vino);
  95. if (!inode) {
  96. struct ceph_mds_request *req;
  97. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
  98. USE_ANY_MDS);
  99. if (IS_ERR(req))
  100. return ERR_CAST(req);
  101. req->r_ino1 = vino;
  102. req->r_num_caps = 1;
  103. err = ceph_mdsc_do_request(mdsc, NULL, req);
  104. inode = req->r_target_inode;
  105. if (inode)
  106. ihold(inode);
  107. ceph_mdsc_put_request(req);
  108. if (!inode)
  109. return ERR_PTR(-ESTALE);
  110. }
  111. dentry = d_obtain_alias(inode);
  112. if (IS_ERR(dentry)) {
  113. pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
  114. fh->ino, inode);
  115. iput(inode);
  116. return dentry;
  117. }
  118. err = ceph_init_dentry(dentry);
  119. if (err < 0) {
  120. iput(inode);
  121. return ERR_PTR(err);
  122. }
  123. dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
  124. return dentry;
  125. }
  126. /*
  127. * convert connectable fh to dentry
  128. */
  129. static struct dentry *__cfh_to_dentry(struct super_block *sb,
  130. struct ceph_nfs_confh *cfh)
  131. {
  132. struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
  133. struct inode *inode;
  134. struct dentry *dentry;
  135. struct ceph_vino vino;
  136. int err;
  137. dout("__cfh_to_dentry %llx (%llx/%x)\n",
  138. cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
  139. vino.ino = cfh->ino;
  140. vino.snap = CEPH_NOSNAP;
  141. inode = ceph_find_inode(sb, vino);
  142. if (!inode) {
  143. struct ceph_mds_request *req;
  144. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
  145. USE_ANY_MDS);
  146. if (IS_ERR(req))
  147. return ERR_CAST(req);
  148. req->r_ino1 = vino;
  149. req->r_ino2.ino = cfh->parent_ino;
  150. req->r_ino2.snap = CEPH_NOSNAP;
  151. req->r_path2 = kmalloc(16, GFP_NOFS);
  152. snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
  153. req->r_num_caps = 1;
  154. err = ceph_mdsc_do_request(mdsc, NULL, req);
  155. inode = req->r_target_inode;
  156. if (inode)
  157. ihold(inode);
  158. ceph_mdsc_put_request(req);
  159. if (!inode)
  160. return ERR_PTR(err ? err : -ESTALE);
  161. }
  162. dentry = d_obtain_alias(inode);
  163. if (IS_ERR(dentry)) {
  164. pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
  165. cfh->ino, inode);
  166. iput(inode);
  167. return dentry;
  168. }
  169. err = ceph_init_dentry(dentry);
  170. if (err < 0) {
  171. iput(inode);
  172. return ERR_PTR(err);
  173. }
  174. dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
  175. return dentry;
  176. }
  177. static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
  178. int fh_len, int fh_type)
  179. {
  180. if (fh_type == 1)
  181. return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw);
  182. else
  183. return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw);
  184. }
  185. /*
  186. * get parent, if possible.
  187. *
  188. * FIXME: we could do better by querying the mds to discover the
  189. * parent.
  190. */
  191. static struct dentry *ceph_fh_to_parent(struct super_block *sb,
  192. struct fid *fid,
  193. int fh_len, int fh_type)
  194. {
  195. struct ceph_nfs_confh *cfh = (void *)fid->raw;
  196. struct ceph_vino vino;
  197. struct inode *inode;
  198. struct dentry *dentry;
  199. int err;
  200. if (fh_type == 1)
  201. return ERR_PTR(-ESTALE);
  202. pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino,
  203. cfh->parent_name_hash);
  204. vino.ino = cfh->ino;
  205. vino.snap = CEPH_NOSNAP;
  206. inode = ceph_find_inode(sb, vino);
  207. if (!inode)
  208. return ERR_PTR(-ESTALE);
  209. dentry = d_obtain_alias(inode);
  210. if (IS_ERR(dentry)) {
  211. pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n",
  212. cfh->ino, inode);
  213. iput(inode);
  214. return dentry;
  215. }
  216. err = ceph_init_dentry(dentry);
  217. if (err < 0) {
  218. iput(inode);
  219. return ERR_PTR(err);
  220. }
  221. dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry);
  222. return dentry;
  223. }
  224. const struct export_operations ceph_export_ops = {
  225. .encode_fh = ceph_encode_fh,
  226. .fh_to_dentry = ceph_fh_to_dentry,
  227. .fh_to_parent = ceph_fh_to_parent,
  228. };